search

AminHP / gym-anytrading

The most simple, flexible, and comprehensive OpenAI Gym trading environment (Approved by OpenAI Gym)
MIT License
2.1k stars 465 forks source link

Problems #29

Closed lorrp1 closed 3 years ago

lorrp1 commented 3 years ago

A

AminHP commented 3 years ago

It seems there is a problem with stable_baselines. I tried to implement a DQN from scratch, and it works just fine with the sin function.

toksis commented 3 years ago

Can you change the learning rate? 

model = DQN(MlpPolicy, env, verbose=1,learning_rate=0.00001)?

I tried to do a learning rate of 1. and it do this.

image

toksis commented 3 years ago

Learning rate of .0001

image

toksis commented 3 years ago

Im using the old stable baseline.

toksis commented 3 years ago

Maybe you can share your sin function code? @lorrp1

lorrp1 commented 3 years ago

@AminHP im getting even worse results with ray

AminHP commented 3 years ago

I suggest you read this article and implement your own DQN and test it with the environment.

toksis commented 3 years ago

try the zoo in stable baselines.

lorrp1 commented 3 years ago

@toksis are you able to get it working with the sin function there?

toksis commented 3 years ago

Sadly no... I want to try the zoo in stable baselines maybe it will help. But first I will try the article from AdminHP.